← LibraryWhen Are Two Twisted Modules Isomorphic? | KEVOS® MathematicsProject Delivery · Project ManagementLesson 69/72← Prev Next →

ArticlePublished 9 Aug 202623 min readBy Kevin Jogin

Skip to content

Engineering MathematicsAdvanced

When Are Two Twisted Modules Isomorphic?

Twisting $K[X]$ by the automorphism $∂_i ↦ ∂_i + g_i$ produces a simple $A_n$ -module for every integrable $g$ , and two such twists are isomorphic exactly when the vectors $g$ coincide. This yields an infinite family of pairwise non-isomorphic simple modules over $A_n$ .

Collection Algebraic D-modulesTopic stream weyl-modulesSource Ch. 5 §2Reading time 27 minPage ID KVS-ENG-MATH-0353

Overview

Twisting takes a left $A_n$ -module $M$ and an automorphism $σ$ of $A_n$ and returns a new module $M_σ$ : the same abelian group, with $a ∈ A_n$ now acting the way $σ(a)$ used to act. The construction is cheap, it preserves simplicity, and $A_n$ has a very large automorphism group. So it is a factory for producing modules. The question this page settles is whether the factory produces anything genuinely new, and how one tells two of its products apart.

The answer, for the family that matters most, is completely clean. Fix polynomials $g_1, … ,g_n ∈ K[X]$ satisfying the integrability condition $∂_i g_j = ∂_j g_i$ , and let $σ_g$ be the automorphism with $σ_g(x_i) = x_i$ and $σ_g(∂_i) = ∂_i + g_i$ . Then $K[X]_σ_g$ is simple, and

K[X]_σ_g ≅ K[X]_σ_h ⟺ g = h.

(5.1)

Different data give non-isomorphic modules, with no exceptions and no coincidences. Coutinho proves the special case $g_i = -x_i^,r$ , $r = 1, 2, 3, …$ , which already gives an infinite family of pairwise non-isomorphic simple $A_n$ -modules; (5.1) upgrades that to a family indexed by all integrable $g$ , which over $ℂ$ is uncountable.

Two mechanisms drive the proof, and both are worth carrying away. First, a homomorphism out of a cyclic module is determined by the image of the generator, so an isomorphism question becomes a differential equation for a single polynomial. Second, a nonzero map between simple modules is automatically an isomorphism, so a non-surjective candidate map is a contradiction rather than a partial result. Beyond this family the isomorphism problem is hard: classifying all simple $A_1$ -modules is a genuinely difficult piece of ring theory, and twisting sees only a corner of it.

Definition

Throughout, $K$ is a field of characteristic zero, $K[X] = K[x_1, … ,x_n]$ , and $A_n$ acts on $K[X]$ in the standard way: $x_i$ multiplies, $∂_i$ differentiates. See the polynomial module for that action and its simplicity.

Twist by an automorphismCoutinho, Ch. 5 §2

Let $R$ be a ring, $M$ a left $R$ -module and $σ$ an automorphism of $R$ . The twisted module $M_σ$ has the same underlying abelian group as $M$ , with the action

a ∗ u = σ(a),u (a ∈ R, u ∈ M).

Associativity holds because $σ$ is multiplicative: $(ab) ∗ u = σ(a)σ(b)u = a ∗ (b ∗ u)$ .

The exponential twists of the polynomial module

Let $g = (g_1, … ,g_n)$ with $g_i ∈ K[X]$ and $∂_i g_j = ∂_j g_i$ for all $i, j$ . The assignments $x_i ↦ x_i$ , $∂_i ↦ ∂_i + g_i$ preserve all the defining relations of $A_n$ — the integrability condition is exactly $[∂_i + g_i, ∂_j + g_j] = 0$ — and therefore define an endomorphism $σ_g$ , which is an automorphism with inverse $σ_-g$ . Write $E_g = K[X]_σ_g$ .

If each $g_i$ depends only on $x_i$ , integrability is automatic; that is the case Coutinho uses.

Twisting a quotientCoutinho (5.2.1)(4)

For a left ideal $J$ of $R$ and an automorphism $σ$ , $(R/J)_σ ≅ R/σ^-1(J)$ . Indeed the map $R → (R/J)_σ$ sending $b ↦ b ∗ 1 = σ(b)$ is a surjective homomorphism of left $R$ -modules with kernel $b : σ(b) ∈ J = σ^-1(J)$ .

The inverse is not decoration

It is $σ^-1(J)$ , not $σ(J)$ . The two choices produce genuinely different modules: for $σ_g$ one gets $A_n/ ∑ _i A_n(∂_i - g_i)$ , whereas the wrong version returns $A_n/ ∑ _i A_n(∂_i + g_i)$ , and by (5.1) those two are not isomorphic unless $g = 0$ .

Core Concepts

Twisting is an action of the automorphism group on isomorphism classes

A direct computation gives $(M_σ)_τ = M_στ$ and $M_id = M$ , and an isomorphism $M \to N$ is still an isomorphism $M_σ → N_σ$ . So twisting is a right action of $Aut(A_n)$ on the set of isomorphism classes of $A_n$ -modules. Consequently

M_σ ≅ M_τ ⟺ M_στ^-1 ≅ M,

(5.2)

and the whole isomorphism problem for twists of a fixed $M$ collapses to the computation of one subgroup: the stabiliser $σ : M_σ ≅ M$ .

Inner automorphisms never help

If $σ(a) = uau^-1$ for a unit $u$ , then $m \mapsto u m$ is an isomorphism $M → M_σ$ , so inner automorphisms lie in every stabiliser. For the Weyl algebra this observation is empty: because Bernstein degree is additive, the units of $A_n$ are the nonzero scalars, so the only inner automorphism is the identity. Every automorphism of $A_n$ is outer, and none of them is automatically trivial on isomorphism classes. Which of them act trivially has to be decided case by case.

The twist of $K[X]$ is a differential equation

The module $E_g = A_n/ ∑ _i A_n(∂_i - g_i)$ is the D-module of the system $∂_i f = g_i f$ . Its formal solution is the exponential $! ( ∈ t g )$ , which is a genuine function but not a polynomial. Twisting by $σ_g$ is therefore the algebraic shadow of the gauge transformation $f ↦ e^ ∈ t g f$ : it multiplies solutions by an exponential factor without changing the underlying vector space. The isomorphism question becomes: when do two exponential factors differ by a polynomial factor? The answer, over a polynomial ring, is never — unless they are equal.

Construction and Proof

Isomorphism criterion for exponential twists

Let $g, h$ be integrable vectors of polynomials. Then every $A_n$ -homomorphism $ϕ : E_g → E_h$ has the form $ϕ(p) = pf$ for a unique $f ∈ K[X]$ satisfying (5.4), and conversely. Consequently $E_g ≅ E_h$ if and only if $g = h$ , and $End_A_n(E_g) = K$ .

Proof

Step 1: a homomorphism is multiplication by a polynomial. In both modules $x_i$ acts by ordinary multiplication, because $σ_g(x_i) = x_i$ . The element $1$ generates $E_g$ , and for $p ∈ K[X]$ we have $p ∗ 1 = p$ . So $ϕ$ is determined by $f := ϕ(1)$ , and $ϕ(p) = ϕ(p ∗ 1) = p ∗ f = pf$ .

Step 2: equivariance for $∂_i$ is equation (5.4). On the one hand $∂_i ∗ 1 = ∂_i(1) + g_i = g_i$ in $E_g$ , so $ϕ(∂_i ∗ 1) = g_i f$ . On the other hand $∂_i ∗ f = ∂_i f + h_i f$ in $E_h$ . Equating gives $∂_i f = (g_i - h_i) f$ . The computation reverses: if $f$ satisfies these equations then $p \mapsto p f$ is $A_n$ -linear, since $∂_i ∗ (pf) = (∂_i p) f + p,∂_i f + h_i pf = (∂_i p)f + g_i p f = ϕ(∂_i ∗ p)$ .

Step 3: a nonzero $f$ forces $f$ to be a constant. Both $E_g$ and $E_h$ are simple, being twists of the simple module $K[X]$ . A nonzero homomorphism between simple modules is an isomorphism, so multiplication by $f$ must be surjective onto $K[X]$ . Multiplication by $f$ has image $fK[X]$ , which is all of $K[X]$ only if $f$ is a unit of $K[X]$ , that is a nonzero scalar.

Step 4: conclude. If $f ∈ K^ ×$ then $∂_i f = 0$ , so (5.4) gives $(g_i - h_i)f = 0$ and hence $g_i = h_i$ for every $i$ , because $K[X]$ is a domain. Conversely $g = h$ gives the identity map. Taking $g = h$ in Steps 1–4 shows every endomorphism is multiplication by a scalar.

A proof by degree, without simplicity

Step 3 can be replaced by a bare degree count, which is how Coutinho argues (5.2.3). Suppose $f ≠ 0$ satisfies (5.4) and $g_i ≠ h_i$ for some $i$ . If $_x_i f = 0$ then $∂_i f = 0$ , so $(g_i - h_i)f = 0$ and the domain property gives $g_i = h_i$ . If $_x_i f ≥ 1$ then $_x_i(∂_i f) = _x_i f - 1$ , while $_x_i! ((g_i - h_i)f ) ≥ _x_i f$ . The two sides of (5.4) cannot agree. The degree version is what makes the argument survive when one only knows a nonzero map exists.

An infinite family of simple modulesCoutinho (5.2.3)

The modules $K[X]_σ_r$ , $r = 1,2,3, …$ , where $σ_r(x_i) = x_i$ and $σ_r(∂_i) = ∂_i - x_i^,r$ , are simple and pairwise non-isomorphic. Over $K = C$ the larger family $E_g$ is uncountable, so $A_n$ has uncountably many pairwise non-isomorphic simple modules even though it is a simple Noetherian domain.

Key Equations

The presentation of an exponential twist, obtained by applying the twisting-a-quotient proposition to $K[X] ≅ A_n/ ∑ _i A_n∂_i$ and $σ_g^-1(∂_i) = ∂_i - g_i$ :

E_g ;=; K[X]_σ_g ; ≅ ; A_n / ∑ _i=1^n A_n(∂_i - g_i).

(5.3)

The equation that any homomorphism $E_g → E_h$ must satisfy, where $f$ is the image of the generator:

(∂ f)/(∂ x_i) = (g_i - h_i),f (i = 1, … ,n).

(5.4)

Coutinho's family is $g_i = -x_i^,r$ ; for $σ_r := σ_(-x_1^r, … ,-x_n^r)$ and $r < t$ , equation (5.4) reads

(∂ f)/(∂ x_i) = (x_i^,t - x_i^,r),f,

(5.5)

which has no nonzero polynomial solution, because the left side has strictly smaller degree in $x_i$ than the right.

For $n = 1$ the twists are separated by a numerical invariant as well. With the filtration induced by the generator,

_K Γ_m = rm + 1, d(K[x]_σ_r) = 1, e(K[x]_σ_r) = r.

(5.6)

Variable Definitions

$K$: the ground field, of characteristic zero
$A_n$: the $n$ -th Weyl algebra over $K$ , with generators $x_1, … ,x_n,∂_1, … ,∂_n$
$K[X]$: the polynomial ring $K[x_1, … ,x_n]$ , viewed as a left $A_n$ -module
$K[∂]$: the polynomial ring in the $∂_i$ , viewed as $A_n/ ∑ _i A_n x_i$
$σ, τ$: automorphisms of $A_n$
$M_σ$: the twist of $M$ by $σ$ , with action $a ∗ u = σ(a)u$
$g = (g_1, … ,g_n)$: a vector of polynomials with $∂_i g_j = ∂_j g_i$
$σ_g$: the automorphism $x_i ↦ x_i$ , $∂_i ↦ ∂_i + g_i$
$E_g$: the twisted module $K[X]_σ_g$
$ℱ$: the Fourier transform automorphism, $x_i ↦ ∂_i$ , $∂_i ↦ -x_i$

Properties and Behaviour

What twisting preservesCoutinho (5.2.1)

For any ring $R$ , automorphism $σ$ and left $R$ -module $M$ : $M_σ$ is simple if and only if $M$ is; $M_σ$ is a torsion module if and only if $M$ is; and $(M/N)_σ = M_σ/N_σ$ for every submodule $N$ . In fact $M ↦ M_σ$ is an equivalence of the category of left $R$ -modules with itself, so every purely categorical property — length, indecomposability, projectivity, the lattice of submodules, $Ext$ groups — is preserved.

Dimension is a twist invariant; multiplicity is not

Let $σ$ be an automorphism of $A_n$ and let $c$ be the largest Bernstein degree of $σ(x_1), … ,σ(∂_n)$ , so that $σ(B_m) ⊂ eq B_cm$ . If $M$ is finitely generated with generators $u_1, … ,u_k$ , the filtration $Γ'_m = B_m ∗ U$ on $M_σ$ satisfies $Γ'_m ⊂ eq Γ_cm$ , where $Γ$ is the corresponding filtration on $M$ and $U$ is the span of the generators. Hence $_K Γ'_m ≤ χ_Γ(cm)$ , a polynomial in $m$ of degree $d(M)$ , giving $d(M_σ) ≤ d(M)$ ; applying the same argument to $σ^-1$ gives equality. The multiplicity only obeys the weaker bound $e(M_σ) ≤ c^,de(M)$ , and (5.6) shows it is genuinely not invariant: $e(K[x]) = 1$ while $e(K[x]_σ_r) = r$ , with $c = r$ , so the bound is sharp.

In particular twisting preserves holonomicity, which is why all the modules $E_g$ are holonomic — a family of holonomic modules as large as the family of integrable $g$ .

The Fourier transform moves $K[X]$ Coutinho (5.2.2)

Let $ℱ$ be the automorphism $x_i ↦ ∂_i$ , $∂_i ↦ -x_i$ (an automorphism because $[-x_i, ∂_j] = δ_ij$ ). Then $F^-1(∂_i) = x_i$ , so $K[X]_F ≅ A_n/ ∑ _i A_n x_i = K[∂]$ . Moreover $K[∂] ≇ K[X]$ : in $K[X]$ the element $1$ is nonzero and killed by every $∂_i$ , whereas in $K[∂]$ the operator $∂_1$ acts by multiplication in a domain and so kills nothing nonzero.

Since $F^2$ sends $x_i ↦ -x_i$ and $∂_i ↦ -∂_i$ , we get $F^-2( ∑ _i A_n∂_i) = ∑ _i A_n ∂_i$ and hence $K[X]_F^2 ≅ K[X]$ . The orbit of $[K[X]]$ under the cyclic group generated by $ℱ$ has exactly two elements, $[K[X]]$ and $[K[∂]]$ .

Some automorphisms of $A_n$ and their effect on the class of $K[X]$ .
Automorphism	Action on generators	$K[X]_σ$	Class moved?
Translation	$x_i ↦ x_i + c_i$ , $∂_i ↦ ∂_i$	$K[X]$	no
Scaling	$x_i ↦ λ x_i$ , $∂_i ↦ λ^-1∂_i$	$K[X]$	no
Shear in $∂$	$x_i ↦ x_i + p_i(∂)$ , $∂_i ↦ ∂_i$	$K[X]$	no
Exponential twist $σ_g$	$x_i ↦ x_i$ , $∂_i ↦ ∂_i + g_i$	$A_n/ ∑ _i A_n(∂_i-g_i)$	yes, unless $g=0$
Fourier $ℱ$	$x_i ↦ ∂_i$ , $∂_i ↦ -x_i$	$K[∂]$	yes
$F^2$	$x_i ↦ -x_i$ , $∂_i ↦ -∂_i$	$K[X]$	no

Examples and Special Cases

The first twist, $r = 1$

For $n = 1$ and $g = -x$ the module is $A_1/A_1(∂ + x)$ , whose analytic solution is $(-x^2/2)$ . It is simple, holonomic, and has $d = e = 1$ , exactly like $K[x]$ — yet it is not isomorphic to $K[x]$ , because $∂ f = -xf$ has no nonzero polynomial solution. This is the smallest case where the numerical invariants fail and equation (5.4) is needed.

Twists that change nothing

Take $g_i = 0$ but $σ$ a translation $x_i ↦ x_i + c_i$ , $∂_i ↦ ∂_i$ . Then $σ^-1$ fixes each $∂_i$ , so $σ^-1( ∑ _i A_n∂_i) = ∑ _i A_n∂_i$ and $K[X]_σ ≅ K[X]$ . Concretely, translation of the variable permutes polynomials without disturbing the flat vector $1$ . A nontrivial automorphism can therefore lie in the stabiliser: outer does not mean class-moving.

Two twists in two variables

In $A_2$ take $g = (x_2, x_1)$ , which is integrable since $∂_2 x_2 = 1 = ∂_1 x_1$ ; the corresponding module is $A_2/(A_2(∂_1 - x_2) + A_2(∂_2 - x_1))$ , with solution $(x_1x_2)$ . Take $h = (0,0)$ , giving $K[X]$ . By (5.1) they are not isomorphic; equation (5.4) here reads $∂_1 f = x_2 f$ and $∂_2 f = x_1 f$ , which no nonzero polynomial satisfies.

The vector $(x_2, x_1^2)$ is not integrable ( $∂_2 x_2 = 1 ≠ 0 = ∂_1 x_1^2$ ), so it defines no automorphism; the corresponding system $∂_1 f = x_2 f$ , $∂_2 f = x_1^2 f$ is inconsistent, and the module $A_2/(A_2(∂_1-x_2)+A_2(∂_2-x_1^2))$ is not a twist of $K[X]$ .

The Dirac module under Fourier transform

The delta module is $A_1/A_1x ≅ K[∂]$ . Since $F^-1(x) = -∂$ , its Fourier twist is $A_1/A_1∂ ≅ K[x]$ . Twisting therefore exchanges the two standard simple modules, matching the analytic statement that the Fourier transform of the delta distribution is a constant.

Worked Example

Separating the twists numerically: $e(K[x]_σ_r) = r$

Step 1 - fix the module and its action
Take $n = 1$ and $M = K[x]_σ_r$ with $σ_r(x) = x$ , $σ_r(∂) = ∂ - x^r$ and $r ≥ 1$ . The underlying space is $K[x]$ ; the action is

$x ∗ f = xf, ∂ ∗ f = f' - x^r f.$

By (5.3), $M ≅ A_1/A_1(∂ + x^r)$ . Filter $M$ by $Γ_m = B_m ∗ 1$ , where $B_m$ is the Bernstein filtration; this is a good filtration because it is generated by the single generator $1$ .
Step 2 - compute the first pieces for $r = 2$
$Γ_0 = K$ . For $Γ_1$ : the spanning set of $B_1$ is $1, x, ∂$ , acting on $1$ to give $1$ , $x$ , and $∂ ∗ 1 = 0 - x^2 = -x^2$ . So $Γ_1 = K ⋅ 1 + Kx + Kx^2$ and $_K Γ_1 = 3$ .

For $Γ_2$ : $B_2$ adds $x^2, x∂, ∂^2$ . We get $x ∗ Γ_1 = Kx + Kx^2 + Kx^3$ and $∂ ∗ (-x^2) = -2x + x^4$ , so $x^4$ enters. Collecting, $Γ_2 = K ⋅ 1 + Kx + Kx^2 + Kx^3 + Kx^4$ and $_K Γ_2 = 5$ . The pattern $1, 3, 5$ matches $2m+1$ .
Step 3 - prove $Γ_m$ is the space of polynomials of degree at most $r m$
Upper bound: multiplying by $x$ raises degree by $1 ≤ r$ , and $f ↦ f' - x^r f$ raises it by at most $r$ , so $Γ_m$ lies in the polynomials of degree $≤ rm$ .

Lower bound, by induction. Suppose $Γ_m$ contains every polynomial of degree $≤ rm$ . Then $Γ_m+1 ⊃ eq Γ_m + xΓ_m$ contains everything of degree $≤ rm+1$ , and for $0 ≤ j ≤ rm$ we have $∂ ∗ x^j = jx^j-1 - x^r+j ∈ Γ_m+1$ , whence $x^r+j ∈ Γ_m+1$ for all such $j$ . Those exponents cover $r, r+1, … , r+rm$ , so $Γ_m+1$ contains everything of degree $≤ r(m+1)$ .

$_K Γ_m = rm + 1 (m ≥ 0).$
Step 4 - read off dimension and multiplicity
The Hilbert polynomial is $χ(m) = rm+1$ , of degree $1$ . So $d(M) = 1 = n$ : every $E_g$ is holonomic. The leading coefficient is $r$ , so $e(M) = 1! ⋅ r = r$ .

Cross-check against the presentation: for $0 ≠ P ∈ A_1$ of Bernstein degree $d$ one has $e(A_1/A_1P) = d$ , and $∂ + x^r$ has Bernstein degree $(1,r) = r$ . The two computations agree.
Step 5 - conclude
Multiplicity is an isomorphism invariant, so $K[x]_σ_r ≅ K[x]_σ_t$ would force $r = t$ . This reproves the corollary for $n = 1$ without touching equation (5.5), and it does more: it exhibits an explicit invariant that separates the members of the family.

Result

$K[x]_σ_r$ is holonomic with $d = 1$ and $e = r$ . In particular $r = 1$ gives $e = 1$ , the same multiplicity as $K[x]$ itself, so multiplicity alone does not separate $K[x]_σ_1$ from $K[x]$ — for that pair one still needs equation (5.4), which gives $∂ f = -xf$ and hence no nonzero polynomial solution.

Applications and Industry Use

In a mathematics topic, this section covers downstream use inside mathematics, computing and engineering rather than a manufactured product.

Exponential factors and irregular singularities. Twisting by $σ_g$ is the algebraic form of multiplying solutions by $( ∈ t g)$ . In the analytic theory of irregular connections, exactly these exponential factors are the discrete data attached to a singular point, and (5.1) is the algebraic reason they cannot be absorbed by a polynomial change of variable.
Supply of test objects. Every construction in the theory — dimension, characteristic variety, inverse and direct images — needs a stock of modules on which to be tested. The family $E_g$ is cheap, explicit, holonomic, and has adjustable multiplicity, which makes it the natural stress test.
Fourier methods. Realising the Fourier transform as a twist converts constant-coefficient operators into polynomials and back. This is what makes the module-theoretic proof of results about constant-coefficient equations short, and it is used in the same way in the theory of the Fourier–Laplace transform of D-modules.
Counterexample construction. Because the family is uncountable while $A_n$ is countable when $K$ is, the twists show that the set of isomorphism classes of simple $A_n$ -modules is strictly larger than any list of finitely presented normal forms one might hope for. See the catalogue of counterexamples.

Computational Notes

Read this as the manufacturing section of the template: how the object is actually built by machine, at what cost, and where the computation stops being decidable.

Deciding $A_n/I ≅ A_n/J$ is an isomorphism test between two cyclic presentations, and it reduces to a $Hom$ computation:

Hom_A_n(A_n/I,, A_n/J) ; ≅ ; , a ∈ A_n/J ;:; Ia ⊂ eq J ,.

Compute Gröbner bases of $I$ and $J$ in the Weyl algebra with respect to a term order refining the Bernstein filtration.
For increasing degree bounds $m$ , solve the linear system over $K$ describing $a ∈ B_m : Ia ⊂ eq J$ modulo $J$ .
For holonomic modules the answer stabilises at a computable bound, so the procedure terminates; a nonzero solution $a$ gives a candidate map, and one checks surjectivity and injectivity by comparing multiplicities.
Cheap pre-filter: compute $d$ and $e$ for both modules first. Different multiplicities settle the question immediately and cost far less than the Hom computation.

The Dmodules package in Macaulay2 implements homomorphism and Ext computations between holonomic modules following Tsai and Walther; dmod.lib in Singular and the ore_algebra package in SageMath provide the underlying Gröbner engines. Note that a negative answer from a degree-bounded search is only conclusive once the bound is justified — an unbounded search that finds nothing proves nothing.

Limits of Validity

The clean criterion (5.1) is a statement about one family, not a general solution to the isomorphism problem.

It applies to twists of $K[X]$ by automorphisms fixing all the $x_i$ . For a general pair $σ, τ$ the reduction (5.2) still holds, but computing the stabiliser of $[K[X]]$ inside $Aut(A_n)$ is not settled by anything on this page.
Characteristic zero is required. In characteristic $p$ the module $K[X]$ is not simple — the subspace spanned by the $p$ -th powers is a submodule — so Step 3 of the proof collapses, and $A_n$ itself is no longer simple. See the positive characteristic page.
Integrability cannot be dropped. If $∂_i g_j ≠ ∂_j g_i$ then $σ_g$ is not an endomorphism of $A_n$ at all, and the module $A_n/ ∑ _i A_n(∂_i - g_i)$ is something else entirely — it can even be zero.
The family does not exhaust the simple modules. Classifying simple $A_1$ -modules is a hard open-ended problem; Block's 1981 work reduces it to conjugacy classes of certain irreducible elements, and the exponential twists are only one visible slice of it.

What the criterion does not say

It says nothing about isomorphisms of $E_g$ with modules constructed some other way. For instance the localisation $K[x][1/x]$ is holonomic of multiplicity $2$ but is not simple, so it is not an $E_g$ ; ruling out such coincidences in general requires invariants beyond $(d,e)$ , such as the characteristic variety or the lattice of submodules.

Failure Modes and Common Mistakes

Assuming a cyclic presentation is unique

A cyclic module is $A_n/I$ for many different left ideals $I$ : the ideal depends on the generator chosen. Take $M = K[x]$ over $A_1$ . With the generator $1$ we get $I = A_1∂$ . With the generator $x$ — which also generates, since $M$ is simple — we get $ann(x) = A_1(x∂ - 1) + A_1∂^2$ , and this ideal is not principal: if it were $A_1P$ , then $e(A_1/A_1P) = P$ would have to equal $e(K[x]) = 1$ , forcing $P = ax + b∂ + c$ , and no such operator annihilates $x$ . So $I ≠ J$ tells you nothing; only similarity of the ideals does.

Sign errors in the twist, and their cost

The two conventions $a ∗ u = σ(a)u$ and $a ∗ u = σ^-1(a)u$ both define modules, and both appear in the literature. Under the first, $K[X]_σ_g ≅ A_n/ ∑ _i A_n(∂_i - g_i)$ ; under the second, the sign of $g$ flips. By (5.1) the two answers are non-isomorphic modules whenever $g ≠ 0$ , so this is not a cosmetic ambiguity. State the convention before computing, and check it against the case $g_i = 0$ .

Reading "simple ring" as "few simple modules"

P0 is a simple ring, so every nonzero module is faithful and the two-sided annihilator carries no information whatsoever about a module. That is the opposite of a classification: it means the cheap invariant is useless, and the family $E_g$ shows there are uncountably many isomorphism classes of simple modules to tell apart.

Expecting numerical invariants to be complete

$K[x]$ and $K[x]_σ_1$ have the same dimension ( $1$ ), the same multiplicity ( $1$ ), the same length ( $1$ ), and both are simple and holonomic. They are still not isomorphic. Numerical invariants can only ever separate classes, never certify that two modules agree; for that one needs an explicit map, and (5.4) is the equation it must satisfy.

Historical Notes

Twisting by an automorphism is old and general ring theory; its use here follows Dixmier's 1968 study of the algebra $A_1$ , which computed automorphisms, classified certain simple modules, and posed the conjecture that every endomorphism of $A_n$ is an automorphism — see the Dixmier conjecture. The exponential twists appear there in the guise of the modules attached to $∂ - g$ .

The isomorphism problem for simple $A_1$ -modules was studied seriously in the late 1970s and 1980s. Block's 1981 paper reduced the classification to data attached to irreducible elements of a localisation of $A_1$ , and made clear that no finite list of families would do. Stafford's work on right ideals of $A_n$ supplied the companion fact for cyclic presentations: left ideals are classified only up to similarity, and the similarity classes are already rich for $A_1$ .

Coutinho's Primer keeps to the concrete end of this story, proving that the twists $K[X]_σ_r$ are pairwise non-isomorphic by a degree count on one differential equation. The point of the argument, then as now, is that questions about modules over a very large noncommutative ring can be reduced to a single polynomial identity.

Comparison

Behaviour of invariants under twisting by an automorphism $σ$ of $A_n$ .
Invariant	Preserved by twisting?	Witness
Simplicity	yes	twisting is a category equivalence
Length	yes	submodule lattice is preserved
Being a torsion module	yes	$ann_M_σ(u) = σ^-1(ann_M(u))$
Dimension $d(M)$	yes	$σ(B_m) ⊂ eq B_cm$ bounds growth both ways
Holonomicity	yes	immediate from dimension
Multiplicity $e(M)$	no	$e(K[x]) = 1$ , $e(K[x]_σ_r) = r$
Characteristic variety	no	$V(ξ)$ for $K[x]$ , $V(x)$ for $K[x]_σ_r$ , $r ≥ 2$
Isomorphism class	no	$K[X]_F ≅ K[∂] ≇ K[X]$

Key Takeaways

Key points

Twisting is a right action of $Aut(A_n)$ on isomorphism classes, so $M_σ ≅ M_τ$ if and only if $M_στ^-1 ≅ M$ .
For a left ideal $J$ , $(A_n/J)_σ ≅ A_n/σ^-1(J)$ — with the inverse, not $σ$ .
The exponential twist $E_g = K[X]_σ_g$ is the D-module of the system $∂_i f = g_i f$ , and $E_g ≅ E_h$ if and only if $g = h$ .
The proof reduces the isomorphism question to the equation $∂_i f = (g_i - h_i)f$ for a single polynomial $f$ , which forces $f$ to be a nonzero scalar.
Coutinho (5.2.3) is the special case $g_i = -x_i^,r$ ; over $ℂ$ the full family is uncountable, so $A_n$ has uncountably many simple modules.
Dimension and holonomicity survive twisting; multiplicity and the characteristic variety do not — $e(K[x]_σ_r) = r$ .
Numerical invariants can separate classes but never certify an isomorphism; only an explicit map does that.

FAQs

Why does an isomorphism have to be multiplication by a polynomial?

Because in every $E_g$ the variables $x_i$ act by ordinary multiplication, and $1$ generates. A homomorphism is determined by the image $f$ of the generator, and $K[X]$ -linearity then forces $p \mapsto p f$ . Only the $∂_i$ -equivariance carries information, and that is equation (5.4).

Is every

E_g

really simple?

Yes. $K[X]$ is a simple $A_n$ -module, and twisting preserves the submodule lattice exactly, so $E_g$ has no proper nonzero submodules either. Simplicity is what makes Step 3 of the proof work: a nonzero map must be onto.

Could two different automorphisms give the same twisted module?

Certainly — translations, scalings and shears in $∂$ all fix the class of $K[X]$ . The statement (5.1) is only that within the family $σ_g$ distinct parameters give distinct classes. The stabiliser of $[K[X]]$ in the full automorphism group is large.

How do these modules relate to functions?

$E_g$ presents the system $∂_i f = g_i f$ , whose solution is $( ∈ t g)$ . Over $ℂ$ that solution lives in the module of holomorphic functions, where it spans a one-dimensional solution space. The module is the algebraic object; the exponential is what it is about.

Does the multiplicity always distinguish the twists?

No. For $n = 1$ the twist by $σ_1$ has $e = 1$ , the same as $K[x]$ , but the two are not isomorphic. Multiplicity is a genuine invariant and often the cheapest one, but it is far from complete.

What breaks in characteristic

p

?

Simplicity. In characteristic $p$ the polynomials in $x_1^p, … ,x_n^p$ form a proper nonzero submodule of $K[X]$ , since $∂_i(x_i^p) = 0$ . The twists are then no longer simple, and $A_n$ itself has a huge centre, so almost none of the argument survives.

Is there an algorithm that decides isomorphism for arbitrary cyclic modules?

For holonomic modules over $A_n$ there are terminating Hom and Ext algorithms, so isomorphism of two explicitly presented holonomic modules is decidable in principle. For non-holonomic finitely generated modules no such general procedure is available, and the underlying question — similarity of left ideals — is already delicate for $A_1$ .

References

S. C. Coutinho, A Primer of Algebraic D-modules, London Mathematical Society Student Texts 33, Cambridge University Press, 1995 — Ch. 5 §2, results (5.2.1), (5.2.2) and (5.2.3).
J. Dixmier, Sur les algèbres de Weyl, Bulletin de la Société Mathématique de France 96 (1968), 209–242 — automorphisms of $A_1$ and its simple modules.
R. E. Block, The irreducible representations of the Lie algebra sl(2) and of the Weyl algebra, Advances in Mathematics 39 (1981), 69–110.
J. T. Stafford, Endomorphisms of right ideals of the Weyl algebra, Transactions of the American Mathematical Society 299 (1987), 623–639.
J. C. McConnell and J. C. Robson, Noncommutative Noetherian Rings, Graduate Studies in Mathematics 30, revised edition, American Mathematical Society, 2001 — Ch. 1 and Ch. 8, for twisting, similarity of ideals and Gelfand–Kirillov dimension.
J.-E. Björk, Rings of Differential Operators, North-Holland, 1979 — Ch. 1, for filtrations and multiplicity.
H. Tsai and U. Walther, Computing homomorphisms between holonomic D-modules, Journal of Symbolic Computation 32 (2001), 597–617 — the algorithms behind the Dmodules package in Macaulay2.
ISO 80000-2:2019, Quantities and units — Part 2: Mathematics, International Organization for Standardization.

AI Suggested Questions

Verify directly that $σ_g$ preserves the relation $[∂_i, x_j] = δ_ij$ , and show that integrability is exactly what is needed for $[∂_i + g_i, ∂_j + g_j] = 0$ .
Compute the characteristic variety of $A_1/A_1(∂ + x^r)$ for $r = 1, 2, 3$ and explain why the answer changes at $r = 1$ .
Show that the stabiliser of the class of $K[X]$ contains every automorphism fixing all the $∂_i$ .
Determine whether $A_1/A_1(∂ - x)$ and $A_1/A_1(∂ + x)$ are isomorphic, and give the polynomial equation that decides it.
Prove that the units of $A_n$ are exactly the nonzero scalars, using additivity of Bernstein degree.
Compute $e(K[x]_σ_g)$ for $g$ a general polynomial of degree $r$ and compare with the bound $e(M_σ) ≤ c^,de(M)$ .
Find an integrable $g$ in two variables that is not a gradient of a polynomial when $K$ has characteristic $p$ , and explain what goes wrong.

Continue learning

A Roadmap Through Algebraic D-module Theory | KEVOS® MathematicsArticle · Project Management Automorphisms of the Weyl Algebra | KEVOS® MathematicsArticle · Project Management Bernstein's Inequality | KEVOS® MathematicsArticle · Project Management Canonical Form of an Element of the Weyl Algebra | KEVOS® MathematicsArticle · Project Management